[Shoumik Chakravarty (shoumikchakravarty-dev)]

Motivation and Context

Closes #13938
Ollama supports image generation https://ollama.com/blog/image-generation through POST /api/generate, which returns a base64-encoded image field for image models. Semantic Kernel has no Ollama text-to-image service, so these models cannot be used through the TextToImageClientBase abstraction.

Description

Adds OllamaTextToImage, following the connector's existing one-service-per-modality

• ollama_text_to_image.py implementing generate_image() via AsyncClient.generate() and returning the
  decoded image bytes
• OLLAMA_IMAGE_MODEL_ID setting on OllamaSettings, matching the existing chat, text and embedding model IDs.
• OllamaTextToImagePromptExecutionSettings carrying width, height and steps.
• Exports from connectors/ai/ollama/__init__.py.
• Adds unit tests covering initialization, custom host and client, base64 decoding, get_image_content, size settings, the deprecated width/height arguments and their precedence, and the missing-image error path.

No new dependencies: the ollama package SK already depends on exposes image on
GenerateResponse, and width, height and steps on generate().

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible

[Copilot]

🟢 Approval recommended

The implementation and tests are complete; only a minor documentation nit remains.

Pull request overview

Adds Ollama text-to-image support with configurable image generation settings and base64-decoded output.

Changes:

• Adds OllamaTextToImage and public exports.
• Adds image model configuration and generation settings.
• Adds comprehensive unit tests.

File summaries

File	Summary
python/tests/unit/connectors/ai/ollama/services/test_ollama_text_to_image.py	Tests image generation behavior and error handling.
python/tests/unit/connectors/ai/ollama/conftest.py	Adds image model test configuration.
python/semantic_kernel/connectors/ai/ollama/services/ollama_text_to_image.py	Implements Ollama image generation.
python/semantic_kernel/connectors/ai/ollama/ollama_settings.py	Adds the image model setting.
python/semantic_kernel/connectors/ai/ollama/ollama_prompt_execution_settings.py	Adds image size and steps settings.
python/semantic_kernel/connectors/ai/ollama/__init__.py	Exports the new service.

Review details

Suppressed comments (1)

python/semantic_kernel/connectors/ai/ollama/init.py:13

• The connector index in python/semantic_kernel/connectors/ai/README.md lists each provider's public modality service, including the existing OpenAI and Azure text-to-image clients, but does not list this new Ollama client. Add OllamaTextToImage there so the exported service is discoverable and the index stays complete.

from semantic_kernel.connectors.ai.ollama.services.ollama_text_to_image import OllamaTextToImage

• Files reviewed: 6/6 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🔵 Needs a closer look

Fix typed-settings handling and keyword collisions, update the central settings reference, and add SDK-shaped response coverage.

Review details

Suppressed comments (5)

Previously missed (1) — in code that hasn't changed since the last review.

python/semantic_kernel/connectors/ai/ollama/ollama_settings.py:34

• The new OLLAMA_IMAGE_MODEL_ID setting is not documented in the repository's settings reference (python/samples/concepts/setup/ALL_SETTINGS.md), which currently lists the three existing Ollama services but not this new service. Please add the corresponding service/model-id/host row so users following the central setup guide can discover and configure the connector.

python/semantic_kernel/connectors/ai/README.md:51

• The settings reference at python/samples/concepts/setup/ALL_SETTINGS.md is linked from the setup documentation and lists each Ollama service, but this new service and its OLLAMA_IMAGE_MODEL_ID environment variable are missing there. Add the service's ai_model_id/host entry so users can discover the required configuration.

|        | [`OllamaTextToImage`](./ollama/services/ollama_text_to_image.py) |

python/semantic_kernel/connectors/ai/ollama/services/ollama_text_to_image.py:104

• This unconditionally calls from_prompt_execution_settings, so an already-typed settings object is repacked into extension_data before each request. If a caller clears a previously set field (for example, settings.width = None), pack_extension_data() leaves the old value in extension_data and this conversion restores it, meaning the request ignores the caller's current settings. Preserve typed settings (or use the service helper) instead.

            else OllamaTextToImagePromptExecutionSettings.from_prompt_execution_settings(settings)

python/semantic_kernel/connectors/ai/ollama/services/ollama_text_to_image.py:127

• Merging kwargs into options and then supplying model, prompt, and stream explicitly makes any of those valid AsyncClient.generate arguments fail with TypeError: multiple values for keyword argument (for example, generate_image(prompt, stream=False)). Since the method documents kwargs as endpoint arguments, reserve these request-control keys or build one dictionary and apply the fixed values before calling the client.

        options = image_settings.prepare_settings_dict()
        options.update(kwargs)

        response_object = await self.client.generate(

python/tests/unit/connectors/ai/ollama/services/test_ollama_text_to_image.py:58

• All success and missing-image mocks here are mappings, but the Ollama SDK returns a GenerateResponse, so the primary getattr(response_object, "image") path in the new service is never exercised. Add a typed/SDK-shaped response case to verify the integration with the actual response object.

    mock_generate.return_value = {"image": ENCODED_IMAGE}

• Files reviewed: 7/7 changed files
• Comments generated: 0 new
• Review effort level: Lite

[Copilot]

Copilot review overview

🟢 Approval recommended

The reviewed changes have no unresolved blocking issues.

Review effort: Lite
Findings: None

[Shoumik Chakravarty (shoumikchakravarty-dev)]

Addressed Copilot's review comment. Latest Copilot pass is green. The two github-app-auth failures are the repo-wide
fork-PR issue (#14438, fix in #14457)

[Eduard van Valkenburg (eavanvalkenburg)]

Thank you for the contribution. We are not adding new features to Semantic Kernel (SK) at the moment, so I am closing this PR. Please consider creating a feature request issue in microsoft/agent-framework.